关键词: 物体识别, 多方向投影, 空间信息, 样本词典, 词袋模型

Abstract: To alleviate the problem that spatial position information is completely ignored in Bags of Words representation, an optimized method based on multi-direction spatial Bags of Words is proposed. Firstly, the representation of spatial sub region of an image is modeled by spatial pyramid. Then, projection method is applied to local features of image blocks in horizontal, vertical and inclined [±45°], spatial structure information is well expressed in multi-direction. Furthermore, it reduces redundant effects of different object and the dimension of features by means of samples visual codebook. Finally, the proposed method is evaluated on two object databases, the results of comparative experiment show that the proposed algorithm has inspiring performance.

Key words: object recognition, multi-direction projection, spatial information, samples codebook, Bags of Words